Towards Chemically Accurate and Scalable Quantum Simulations on IQM Quantum Hardware: A Quantum-HPC Hybrid Approach

This paper presents a large-scale experimental study on IQM's 24-qubit superconducting processor demonstrating that hybrid quantum-classical approaches, specifically combining Sample-based Quantum Diagonalization with various ansätze and Density Matrix Embedding Theory, can achieve chemically accurate ground-state energies and full potential energy surfaces for molecules ranging from simple benchmarks to pharmacologically relevant systems like amantadine.

The Gist

Imagine you are trying to solve a massive, incredibly complex jigsaw puzzle. This puzzle represents the behavior of electrons in a molecule. In the real world, these electrons are constantly dancing, repelling each other, and forming bonds. To understand how a new medicine works or how to build a better battery, scientists need to solve this puzzle perfectly.

The problem? The puzzle has more pieces than there are atoms in the universe. Even the world's most powerful supercomputers get stuck trying to solve it for anything larger than a tiny molecule. They run out of memory and time.

This paper is about a team of scientists who decided to try a different approach: using a quantum computer to help solve the puzzle.

Here is the story of what they did, explained simply:

1. The Problem: The "Curse of Dimensions"

Think of a molecule like a crowded dance floor. If you have 10 dancers, it's easy to predict where they will be. But if you have 100 dancers, and every single one can move in a million different ways simultaneously, the number of possible dance moves becomes impossible for a human (or a normal computer) to track. This is called the "curse of dimensionality."

2. The Solution: The "Smart Sampler" (SQD)

Instead of trying to calculate every possible dance move (which takes forever), the scientists used a method called Sample-based Quantum Diagonalization (SQD).

• The Analogy: Imagine you want to know the average height of everyone in a stadium. You could measure every single person (impossible), or you could ask a quantum computer to quickly take a "snapshot" of the crowd.
• How it works: The quantum computer doesn't solve the whole puzzle. Instead, it acts like a super-fast camera, taking thousands of snapshots of the electrons to see which "dance moves" (configurations) are the most important.
• The Hybrid Team: Once the quantum camera takes the photos, a classical supercomputer (the "brain") takes those specific photos, arranges them, and solves the math for just those few important moves. This is a Quantum-HPC Hybrid approach: the quantum part finds the clues, and the classical part solves the mystery.

3. The Hardware: The "Star Topology"

They used a specific quantum computer made by a company called IQM, named Sirius.

• The Analogy: Think of this computer as a star-shaped network. There is a central hub (a resonator) that connects to 24 "satellites" (qubits). This allows any satellite to talk to any other satellite instantly, which is crucial for solving these complex puzzles without getting tangled up.

4. The Experiments: From Tiny to Huge

The team tested their method on several levels of difficulty:

• Level 1: The Tiny Molecules (H2, LiH, etc.)
  They started with simple molecules like Hydrogen. They proved their method could predict the energy of these molecules with "chemical accuracy" (meaning it's accurate enough to be useful for real science).

• Level 2: The 1D and 2D Maps (The "Terrain")
  Instead of just looking at one static picture, they mapped out the "terrain" of the molecules.

  • 1D Scan: Imagine stretching a rubber band (a molecule) and measuring the energy at every inch. They did this perfectly.
  • 2D Scan (The Big Win): They created a 32x32 grid for a water molecule. Imagine a topographical map of a mountain, where every point on the map shows the energy. This is the first time anyone has successfully mapped a 2D energy landscape for water on a real quantum computer. It's like drawing a detailed weather map for a tiny storm.

• Level 3: The Big Molecules (Embedding)
  This is the most exciting part. They wanted to simulate big molecules, like Amantadine (a drug used for Parkinson's and the flu).

  • The Problem: The whole molecule is too big for the quantum computer.
  • The Trick (DMET): They used a technique called Density Matrix Embedding Theory. Imagine you want to understand a whole city, but you only have a small telescope. Instead of trying to see the whole city at once, you zoom in on one neighborhood (a "fragment"), study it in high detail with your telescope, and then use math to guess how the rest of the city affects that neighborhood.
  • The Result: They successfully simulated the drug molecule by breaking it into small chunks, solving each chunk on the quantum computer, and stitching the answers together.

5. Two Different "Lenses" (Ansätze)

They tried two different ways to prepare the quantum computer for taking its "photos":

1. LUCJ: A method that is very efficient and works well on noisy hardware. It's like using a lightweight, agile camera.
2. LCNot-UCCSD: A method that is more theoretically perfect but creates very deep, complex circuits. It's like using a heavy, high-end camera that is so sensitive to shaking (noise) that it sometimes fails to take a clear picture on larger molecules.

• Conclusion: The lightweight, agile camera (LUCJ) won the race for now because it was more reliable on current hardware.

The Big Picture: Why Does This Matter?

This paper is a major milestone because it proves that we don't need a perfect, error-free quantum computer to do useful chemistry today.

By combining a noisy quantum computer (which takes "guesses" or samples) with a powerful classical supercomputer (which does the heavy math), they can:

1. Simulate molecules that are too big for normal computers.
2. Get results that are accurate enough to help design new drugs and materials.
3. Map out complex energy landscapes (like the water molecule map) that were previously impossible to see.

In short: They showed that by using a quantum computer as a "smart assistant" rather than a "magic solver," we can start solving real-world chemical problems right now, paving the way for better medicines, cleaner energy, and new materials in the future.

Technical Summary

1. Problem Statement

The primary challenge in computational quantum chemistry is the "curse of dimensionality." Exact solutions to the electronic Schrödinger equation (Full Configuration Interaction or FCI) scale factorially with the number of electrons and orbitals, rendering them intractable for all but the smallest systems on classical supercomputers. While classical approximations like Coupled Cluster (CCSD) offer high accuracy, they fail in strongly correlated regimes (e.g., bond dissociation) and scale poorly ($O(N^7)$) for large systems.

Quantum computing offers a potential solution by representing wavefunctions in a Hilbert space that scales exponentially with qubits. However, current Noisy Intermediate-Scale Quantum (NISQ) devices suffer from limited coherence times and high error rates. Traditional Variational Quantum Eigensolvers (VQE) struggle with "barren plateaus" (flat optimization landscapes) and prohibitive measurement overheads. There is a critical need for robust, scalable quantum algorithms that can deliver chemically accurate results on current hardware without relying on deep, error-prone circuits or extensive variational optimization.

2. Methodology

The authors propose a Quantum-HPC Hybrid approach centered on Sample-based Quantum Diagonalization (SQD). This method decouples the quantum sampling task from the energy optimization loop.

• Core Algorithm (SQD):

  • Quantum Sampling: A quantum processor prepares a trial state and measures it in the computational basis to generate a set of dominant Slater determinants (bitstrings).
  • Classical Post-Processing: The quantum device acts only as a sampler. The classical supercomputer constructs the Hamiltonian matrix within the subspace spanned by these sampled determinants and performs exact diagonalization (using the Davidson method) to obtain the ground-state energy.
  • Error Mitigation: The framework employs Configuration Recovery (CR), a symmetry-aware post-processing step that probabilistically corrects noisy bitstrings to satisfy particle number and spin conservation, recovering useful information even from noisy hardware.

• Ansätze Evaluated:

  1. LUCJ (Local Unitary Cluster Jastrow): A hardware-efficient ansatz inspired by Hubbard models. It uses localized orbital rotations and Jastrow factors. It is initialized with CCSD amplitudes (high classical cost, $O(N^6)$) but yields shallow circuits ($O(N^2)$ scaling).
  2. LCNot-UCCSD (Linear-CNOT Unitary Coupled-Cluster Singles and Doubles): A chemically motivated ansatz using linearly scaling CNOT gates. It is initialized with MP2 amplitudes (lower classical cost, $O(N^4)$) but results in significantly deeper circuits ($O(N^4)$ scaling), making it more susceptible to noise.

• Embedding Strategy (DMET):
  To scale beyond small molecules, the authors integrate SQD with Density Matrix Embedding Theory (DMET). The system is fragmented into smaller "impurities" (active spaces) embedded in a mean-field bath. Each impurity is solved using SQD(LUCJ), and the results are combined self-consistently to recover the total energy of the macroscopic system.

• Hardware: Experiments were conducted on IQM Sirius, a 24-qubit superconducting quantum processor with a star topology allowing all-to-all connectivity via a central resonator. Up to 16 qubits were utilized.

3. Key Contributions

• Comprehensive Benchmarking: One of the most extensive experimental demonstrations of quantum chemistry on superconducting hardware, covering ground-state energies, 1D and 2D Potential Energy Surfaces (PES), and embedding for large molecules.
• First Experimental 2D-PES: The first experimental generation of a full 32×32 two-dimensional potential energy surface (bond length vs. bond angle) for the water molecule ($H_2O$) on a superconducting QPU.
• Novel Ansatz Comparison: A direct hardware comparison between the hardware-efficient LUCJ and the chemically rigorous LCNot-UCCSD, highlighting the trade-off between classical pre-computation costs and quantum circuit depth/noise resilience.
• Scalable Embedding Demonstration: The first experimental application of DMET-SQD to a pharmacologically relevant drug molecule, Amantadine ($C_{10}H_{17}N$, 151 Da), demonstrating the feasibility of simulating systems far beyond the reach of standalone NISQ algorithms.
• Chemical Accuracy: Demonstrated that SQD can achieve results within chemical accuracy (1 kcal/mol or ~1.6 mHa) for a diverse set of molecules, including those with strong static correlation.

4. Key Results

• Small Molecule Benchmarks:

  • Tested on $H_2$, $LiH$, $BeH_2$, $H_2O$, and $NH_3$ using STO-3G and 6-31G basis sets.
  • LUCJ consistently achieved sub-nanohartree accuracy (matching FCI) across all molecules when the subsampling budget was sufficient.
  • LCNot-UCCSD matched LUCJ accuracy for small systems ($H_2$, $LiH$, $BeH_2$) but failed for larger systems ($H_2O$, $NH_3$) due to excessive circuit depth causing noise to overwhelm the signal, preventing valid configuration recovery.
  • Conclusion: LUCJ is the superior choice for near-term hardware due to its noise resilience, despite higher classical initialization costs.

• Potential Energy Surfaces (PES):

  • 1D-PES: Successfully mapped dissociation curves for $H_2$, $HeH^+$, $LiH$, and $BeH_2$. The method correctly captured the breakdown of CCSD in the dissociation limit (strong correlation) where CCSD errors grew to $>10^{-2}$ Ha, while SQD remained accurate.
  • 2D-PES ($H_2O$): Generated a 1,024-point grid ($32 \times 32$) of bond lengths and angles. Under full subsampling ($\epsilon_s = 10^8$), the surface was indistinguishable from FCI. Under reduced subsampling, the surface retained global topology but exhibited stochastic noise near the minimum, highlighting the importance of sampling budget.

• DMET Embedding (Ligands & Amantadine):

  • Ligand-like Molecules: Achieved micro-Hartree accuracy (orders of magnitude better than chemical accuracy) for 8 small ligand molecules compared to DMET-CASCI references.
  • Amantadine: Successfully simulated the 151 Da drug molecule by fragmenting it into 11 impurities. The DMET-SQD energy agreed with the DMET-CASCI reference within micro-Hartree precision, significantly outperforming DMET-CCSD.
  • Shot Efficiency: Found that ~1,000 shots were sufficient to saturate the subspace for 16-qubit impurities, indicating that massive shot counts are not always necessary if the configuration recovery is effective.

5. Significance and Outlook

This work establishes IQM quantum hardware as a viable platform for scalable, chemically accurate quantum simulations. The study validates Sample-based Quantum Diagonalization (SQD) as a robust alternative to VQE, effectively bypassing the variational optimization loop and barren plateaus.

• Hybrid Workflow: The success of the Quantum-HPC hybrid model (QPU for sampling, HPC for diagonalization) demonstrates a practical path forward for the NISQ era.
• Scalability: By combining SQD with DMET, the authors have demonstrated a pathway to simulate macroscopic chemical systems (like drugs) that are currently impossible to treat with exact classical methods.
• Future Directions: The paper identifies that while current results are promising, future improvements require better error mitigation for configuration recovery, multi-reference ansatz initialization, and scaling to larger active spaces on higher-fidelity hardware.

In summary, this paper provides a rigorous experimental validation that sample-based algorithms combined with embedding theories can deliver chemically accurate insights into complex molecular systems using current superconducting quantum processors, bridging the gap between theoretical quantum advantage and practical chemical application.